Thermal throttling originated as a safety feature. The early implementations were basically a "thermal fuse" in function, and cut all power to the system to prevent catastrophic hardware damage. Only later did the more sophisticated versions that do things like "cut down clocks to prevent temps from rising further" appear.

On desktop PCs, thermal throttling is often set up as "just a safety feature" to this very day. Which means: the system does NOT expect to stay at the edge of its thermal limit. I would not trust thermal throttling with keeping a system running safely at a continuous 100C on die.

100C is already a "danger zone", with elevated error rates and faster circuit degradation - and there are only this many thermal sensors a die has. Some under-sensored hotspots may be running a few degrees higher than that. Which may not be enough to kill the die outright - but more than enough to put those hotspots into a "fuck around" zone of increased instability and massively accelerated degradation.

If you're relying on thermal throttling to balance your system's performance, as laptops and smartphones often do, then you seriously need to dial in better temperature thresholds. 100C is way too spicy.